John Allspaw, Operations Engineer at Flickr posted a great presentation yesterday about the unique capacity planning and management needs of web operations. My favorite quip was, "Clouds need planning too".
He argues that you must know with certainty your infrastructure ceilings - the preverbial "out of gas" signal where work starts to degrade or fail. Forget benchmarks according to John. Instead, he offers a few "Stupid Capacity Tricks" (which are actually quite cleaver) to get rid of wandering bottlenecks and to keep the cloud running.
John has a lot more to say on the topic in his blog, the Kitchen Soap, so check him out.
PS: I have to also tip my hat to John for showing the big savings from replacing 26 Dell PE860 servers with 8 ProLiant DL140's. Wait until he gets a load of the ProLiant BL2x220c 2-in-1 server blade!